Table 4 Ablation study results of the ADReSSo Challenge dataset. Values presented are the mean ± standard deviation. Results are averaged over four runs.

From: Alzheimer’s disease recognition using graph neural network by leveraging image-text similarity from vision language model

Architecture

Pr (%)

Rc (%)

F1 (%)

Sp (%)

Ac (%)

Shuffling edge weights

81.70 ±4.55

72.86 ±7.56

76.87 ±5.14

84.03 ±4.74

78.52 ±4.36

Independent embedding

76.52 ±4.67

89.29 ±4.29

82.27 ±2.11

72.92 ±7.31

80.99 ±2.70

Shuffling & Ind. emb.

77.93 ±1.66

90.71 ±2.74

83.82 ±1.71

75.00 ±2.27

82.75 ±1.77

Original picture

84.80 ±3.38

75.00 ±1.43

79.56 ±1.16

86.81 ±3.50

80.99 ±1.41

Max Pooling

89.08 ±3.92

81.43 ±3.69

85.07 ±3.66

90.28 ±3.59

85.92 ±3.45

Proposed model

90.93 ±2.34

85.71 ±2.33

88.23 ±2.09

91.67 ±2.27

88.73 ±1.99

  1. Ac accuracy, F1 F1-score, Pr precision, Rc recal, Sp specificity
Back to article page